Synthesis
The primary outcome was LoS. Secondary outcomes were
adverse events, final CSS scores and rate of readmission.
Data on LoS and final mean CSS was used to calculate
mean differences (MD) with 95 % Confidence Intervals
(CIs) for each outcome. Where trials included arms using
different concentrations of hypertonic saline in addition to
a control arm we treated them as two separate trials, dividing
the control arm numbers by two in order not to
double-count control participants. Meta-analyses were
undertaken in RevMan Version 5.2 and Stata version12
using both fixed (Inverse-Variance) and random effect
(Der Simonian & Laird) models [32]; additional sensitivity
analyses used the metasens package as implemented
within the R software version 3.1.3[33]. The four prespecified
intervention subgroups were separated in the
main forest plot to give subgroup estimates of treatment
effects. Statistical heterogeneity—a measure of withintrial
variation—was investigated using the I2 statistic
[30]: 0–40 % indicating unimportant levels; 30–60 %
showing moderate heterogeneity; 50–100 % as demonstrable
heterogeneity within trials. Sensitivity analyses
were undertaken as proposed by Deschartres et al. [34]
along with meta-regression to assess whether heterogeneity
could be attributed to measurable sources. An assessment
for the potential of publication bias was made via assessment
of the funnel plot generated for all trials included in
the meta-analysis. We produced a narrative description of
adverse events and readmission rates.
Results
Searches and selection
After removal of duplicates, the searches yielded 1489
citations, from which 1443 were excluded at the title/
abstract stage (Fig. 1). We retrieved 46 full papers for
further review of which 28 were excluded because they
were not a controlled trial (n = 4) [35–38], were in the
wrong setting (n = 7) [39–45] or wrong population or
intervention (n = 15) [17, 46–59], or were not available
in English (n = 2) [60, 61]. All included studies were
parallel group, participant-randomised trials. 3 were
multi-centre; 9 were single centre and the number of
centres was unknown for 6. We identified 18 trials (2225
participants, excluding one trial as the number of participants
is unknown [26]) eligible for inclusion at the full
paper review stage (Fig. 1). Details of all excluded trials at
the full stage can be found in Additional file 5.